An Enhanced Neural Network Collaborative Filtering (ENNCF) for Personalized Recommender System

Research and development in recommender systems are relatively vigorous and has benefited by recent advancements in deep learning and artificial intelligence algorithms. By creating individualized predictions, recommender systems have shown to be an effective way to manage information overload. Deep learning-based recommender systems have achieved outstanding results, yet the majority of these systems are conventional recommender systems that use a single rating. In this study, we studied enhanced neural network collaborative filtering (ENNCF) technique to overcome the existing shortcomings especially data sparsity and to improve the performance accuracy. In pre-processing, k-recursive reliability-based missing value imputation has been utilized to handle the missing values. The imputed dataset is fed to the neural architecture to predict the user preferences. The studied model outperformed several baselines in the experiments on standard datasets including MovieLens and Yelp. The experimental results demonstrated the superiority, feasibility, stability, and robustness of the studied model.